1. Field of the Invention
The invention relates generally to control of access to data and relates more specifically to control of access to data in a distributed environment.
2. Description of Related Art
The Internet has revolutionized data communications. It has done so by providing protocols and addressing schemes which make it possible for any computer system anywhere in the world to exchange information with any other computer system anywhere in the world, regardless of the computer system's physical hardware, the kind of physical network it is connected to, or the kinds of physical networks that are used to send the information from the one computer system to the other computer system. All that is required for the two computer systems to exchange information is that each computer system have an Internet address and the software necessary for the protocols and that there be a route between the two machines by way of some combination of the many physical networks that may be used to carry messages constructed according to the protocols.
The very ease with which computer systems may exchange information via the Internet has, however, caused problems. On the one hand, it has made accessing information easier and cheaper than it ever was before; on the other hand, it has made it much harder to protect information. The Internet has made it harder to protect information in two ways:                It is harder to restrict access. If information may be accessed at all via the Internet, it is potentially accessible to anyone with access to the Internet. Once there is Internet access to information, blocking skilled intruders becomes a difficult technical problem.        It is harder to maintain security en route through the Internet. The Internet is implemented as a packet switching network. It is impossible to predict what route a message will take through the network. It is further impossible to ensure the security of all of the switches, or to ensure that the portions of the message, including those which specify its source or destination, have not been read or altered en route.        
FIG. 1 shows techniques presently used to increase security in networks that are accessible via the Internet. FIG. 1 shows network 101, which is made up of two separate internal networks 103(A) and 103(B) that are connected by Internet 111. Networks 103(A) and 103(B) are not generally accessible, but are part of the Internet in the sense that computer systems in these networks have Internet addresses and employ Internet protocols to exchange information. Two such computer systems appear in FIG. 1 as requestor 105 in network 103(A) and server 113 in network 103(b). Requestor 105 is requesting access to data which can be provided by server 113. Attached to server 113 is a mass storage device 115 that contains data 117 which is being requested by requester 105. Of course, for other data, server 113 may be the requester and requestor 105 the server. Moreover, access is to be understood in the present context as any operation which can read or change data stored on server 113 or which can change the state of server 113. In making the request, requestor 105 is using one of the standard TCP/IP protocols. As used here, a protocol is a description of a set of messages that can be used to exchange information between computer systems. The actual messages that are sent between computer systems that are communicating according to a protocol are collectively termed a session. During the session, Requestor 105 sends messages according to the protocol to server 113's Internet address and server 113 sends messages according to the protocol to requester 105's Internet address. Both the request and response will travel between internal network 103(A) and 103(B) by Internet 111. If server 113 permits requestor 105 to access the data, some of the messages flowing from server 113 to requestor 105 in the session will include the requested data 117. The software components of server 113 which respond to the messages as required by the protocol are termed a service.
If the owner of internal networks 103(A and B) wants to be sure that only users of computer systems connected directly to networks 103(A and B) can access data 117 and that the contents of the request and response are not known outside those networks, the owner must solve two problems: making sure that server 113 does not respond to requests from computer systems other than those connected to the internal networks and making sure that people with access to Internet 111 cannot access or modify the request and response while they are in transit through Internet 111. Two techniques which make it possible to achieve these goals are firewalls and tunneling using encryption.
Conceptually, a firewall is a barrier between an internal network and the rest of Internet 111. Firewalls appear at 109(A) and (B). Firewall 109(A) protects internal network 103(A) and firewall 109(B) protects internal network 103 (B). Firewalls are implemented by means of a gateway running in a computer system that is installed at the point where an internal network is connected to the Internet. Included in the gateway is an access filter: a set of software and hardware components in the computer system which checks all requests from outside the internal network for information stored inside the internal network and only sends a request on into the internal network if it is from a sources that has the right to access the information. Otherwise, it discards the request. Two such access filters, access filter 107(A), and access filter 107(B), appear in FIG. 1.
A source has the right to access the requested information if two questions can be answered affirmatively:                Is the source in fact who or what it claims to be?        Does the source have the right to access the data?        
The process of finding the answer to the first question is termed authentication. A user authenticates himself or herself to the firewall by providing information to the firewall that identifies the user. Among such information is the following:                information provided by an authentication token (sometimes called a smart card) in the possession of the user,        the operating system identification for the user's machine; and        the IP address and the Internet domain name of the user's machine.        
The information that the firewall uses for authentication can either be in band, that is, it is part of the protocol, or it can be out of band, that is, it is provided by a separate protocol.
As is clear from the above list of identification information, the degree to which a firewall can trust identification information to authenticate a user depends on the kind of identification information. For example, the IP address in a packet can be changed by anyone who can intercept the packet; consequently, the firewall can put little trust in it and authentication by means of the IP address is said to have a very low trust level On the other hand, when the identification information comes from a token, the firewall can give the identification a much higher trust level, since the token would fail to identify the user only if it had come into someone else's possession. For a discussion on authentication generally, see S. Bellovin and W. Cheswick, Firewalls and Internet Security, Addison Wesley, Reading, Mass., 1994.
In modern access filters, access is checked at two levels, the Internet packet, or IP level, and the application level. Beginning with the IP level, the messages used in Internet protocols are carried in packets called data grams. Each such packet has a header which contains information indicating the source and destination of the packet. The source and destination are each expressed in terms of IP address and port number. A port number is a number from 1 to 65535 used to individuate multiple streams of traffic within a computer. Services for well-known Internet protocols (such as HTTP or FTP) are assigned well known port numbers that they ‘listen’ to. The access filter has a set of rules which indicate which destinations may receive IP packets from which sources, and if the source and destination specified in the header do not conform to these rules, the packet is discarded. For example, the rules may allow or disallow all access from one computer to another, or limit access to a particular service (specified by the port number) based on the source of the IP packet. There is, however, no information in the header of the IP packet about the individual piece of information being accessed and the only information about the user is the source information. Access checking that involves either authentication of the user beyond what is possible using the source information or determining whether the user has access to an individual piece of information thus cannot by done at the IP level, but must instead be done at the protocol level.
Access checking at the application level is usually done in the firewall by proxies. A proxy is a software component of the access filter. The proxy is so called because it serves as the protocol's stand-in in the access filter for the purposes of carrying out user authentication and/or access checking on the piece of information that the user has requested. For example, a frequently-used TCP/IP protocol is the hypertext transfer protocol, or HTTP, which is used to transfer World-Wide Web pages from one computer to another such computer system. If access control for individual pages is needed, the contents of the protocol must be inspected to determine which particular Web page is requested. For a detailed discussion of firewalls, see the Bellovin and Cheswick reference supra.
While properly-done access filtering can prevent unauthorized access via Internet 111 to data stored in an internal network, it cannot prevent unauthorized access to data that is in transit through Internet 111. That is prevented by means of tunneling using encryption. This kind of tunneling works as follows: when access filter 107(A) receives an IP packet from a computer system in internal network 103(A) which has a destination address in internal network 103(B), it encrypts the IP packet, including its header, and adds a new header which specifies the IP address of access filter 107(A) as the source address for the packet and the IP address of access filter 107(B) as the destination address. The new header may also contain authentication information which identifies access filter 107(A) as the source of the encrypted packet and information from which access filter 107(B) can determine whether the encrypted packet has been tampered with.
Because the original IP packet has been encrypted, neither the header nor the contents of the original IP packet can be read while it is passing through Internet 111, nor can the header or data of the original IP packet be modified without detection. When access filter 107(B) receives the IP packet, it uses any identification information to determine whether the packet is really from access filter 107(A). If it is, it removes the header added by access filter 107(A) to the packet, determines whether the packet was tampered with and if it was not, decrypts the packet and performs IP-level access checking on the original header. If the header passes, access filter 107(B) forwards the packet to the IP address in the internal network specified in the original header or to a proxy for protocol level access control. The original IP packet is said to tunnel through Internet 111. In FIG. 1, one such tunnel 112 is shown between access filter 107(A) and 107(B). An additional advantage of tunneling is that it hides the structure of the internal networks from those who have access to them only from Internet 111, since the only unencrypted IP addresses are those of the access filters.
The owner of internal networks 103(A) and 103(B) can also use tunneling together with Internet 111 to make the two internal networks 103(A and B) into a single virtual private network (VPN) 119. By means of tunnel 112, computer systems in network 103(A) and 103(B) can communicate with each other securely and refer to other computers as if network 103(A) and 103(B) were connected by a private physical link instead of by Internet 111. Indeed, virtual private network 119 may be extended to include any user who has access to Internet 111 and can do the following:                encrypt Internet packets addressed to a computer system in an internal network 103 in a fashion which permits an access filter 107 to decrypt them;        add a header to the encrypted packet which is addressed to filter 107; and        authenticate him or herself to access filter 107.        
For example, an employee who has a portable computer that is connected to Internet 111 and has the necessary encryption and authentication capabilities can use the virtual private network to securely retrieve data from a computer system in one of the internal networks.
Once internal networks begin using Internet addressing and Internet protocols and are connected into virtual private networks, the browsers that have been developed for the Internet can be used as well in the internal networks 103, and from the point of view of the user, there is no difference between accessing data in Internet 111 and accessing it in internal network 103. Internal network 103 has thus become an Internet, that is, an internal network that has the same user interface as Internet 111. Of course, once all of the internal networks belonging to an entity have been combined into a single virtual private intranet, the access control issues characteristic of the Internet arise again—except this time with regard to internal access to data. While firewalls at the points where the internal networks are connected to Internet 111 are perfectly sufficient to keep outsiders from accessing data in the internal networks, they cannot keep insiders from accessing that data. For example, it may be just as important to a company to protect its personnel data from its employees as to protect it from outsiders. At the same time, the company may want to make its World Wide Web site on a computer system in one of the internal networks 103 easily accessible to anyone who has access to Internet 111.
One solution to the security problems posed by virtual private intranets is to use firewalls to subdivide the internal networks, as well as to protect the internal networks from unauthorized access via the Internet. Present-day access filters 107 are designed for protecting the perimeter of an internal network from unauthorized access, and there is typically only one access filter 107 per Internet connection. If access filters are to be used within the internal networks, there will be many more of them, and virtual private networks that use multiple present-day access filters 107 are not easily scalable, that is, in virtual private networks with small numbers of access filters, the access filters are not a serious burden; in networks with large numbers of access filters, they are. Among the problems posed by present-day access filters when they are present in large numbers in a virtual private network are the following:                Present-day access filters are designed to be centrally-administered by a small number of data security experts. As the number of access filters increases, central administration becomes too slow, too expensive, and too error-prone.        Present-day access filters are designed on the assumption that there are only a small number of access filters between the source and destination for data. Where there are many, the increase in access time and the reduction in access speed caused by the filters becomes important.        Present-day access filters are designed on the assumption that the Internet side of the filter is completely insecure and the internal network side of the filter is completely secure. In fact, both kinds of networks offer varying degrees of security. Because security adds overhead, the access filter should neither require nor provide more than is necessary.        Present-day access filters, where they use encryption, require that each access filter know encryption keys for each other access filter. Large numbers of access filters require substantial duplicated effort in key maintenance.        Present-day access filters do not provide any mechanism for giving the user a view of the information resources that corresponds to the user's access rights.        
What is needed if intranets and virtual private networks are to achieve their full promise is access filters that do not present the above problems for scalability.